[PATCH 1/2] io_uring/rsrc: keep one global dummy_ubuf

[Pavel Begunkov <asml.silence@xxxxxxxxx>]

We set empty registered buffers to dummy_ubuf as an optimisation.
Currently, we allocate the dummy entry for each ring, whenever we can
simply have one global instance.

Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
---
 io_uring/io_uring.c |  9 ---------
 io_uring/rsrc.c     | 14 ++++++++++----
 2 files changed, 10 insertions(+), 13 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index fb70ae436db6..3c97401240c2 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -307,13 +307,6 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 		goto err;
 	if (io_alloc_hash_table(&ctx->cancel_table_locked, hash_bits))
 		goto err;
-
-	ctx->dummy_ubuf = kzalloc(sizeof(*ctx->dummy_ubuf), GFP_KERNEL);
-	if (!ctx->dummy_ubuf)
-		goto err;
-	/* set invalid range, so io_import_fixed() fails meeting it */
-	ctx->dummy_ubuf->ubuf = -1UL;
-
 	if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
 			    0, GFP_KERNEL))
 		goto err;
@@ -352,7 +345,6 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 	INIT_WQ_LIST(&ctx->submit_state.compl_reqs);
 	return ctx;
 err:
-	kfree(ctx->dummy_ubuf);
 	kfree(ctx->cancel_table.hbs);
 	kfree(ctx->cancel_table_locked.hbs);
 	kfree(ctx->io_bl);
@@ -2905,7 +2897,6 @@ static __cold void io_ring_ctx_free(struct io_ring_ctx *ctx)
 		io_wq_put_hash(ctx->hash_map);
 	kfree(ctx->cancel_table.hbs);
 	kfree(ctx->cancel_table_locked.hbs);
-	kfree(ctx->dummy_ubuf);
 	kfree(ctx->io_bl);
 	xa_destroy(&ctx->io_bl_xa);
 	kfree(ctx);
diff --git a/io_uring/rsrc.c b/io_uring/rsrc.c
index 5e8fdd9b8ca6..92e2471283ba 100644
--- a/io_uring/rsrc.c
+++ b/io_uring/rsrc.c
@@ -33,6 +33,12 @@ static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
 #define IORING_MAX_FIXED_FILES	(1U << 20)
 #define IORING_MAX_REG_BUFFERS	(1U << 14)
 
+static const struct io_mapped_ubuf dummy_ubuf = {
+	/* set invalid range, so io_import_fixed() fails meeting it */
+	.ubuf = -1UL,
+	.ubuf_end = 0,
+};
+
 int __io_account_mem(struct user_struct *user, unsigned long nr_pages)
 {
 	unsigned long page_limit, cur_pages, new_pages;
@@ -132,7 +138,7 @@ static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf **slo
 	struct io_mapped_ubuf *imu = *slot;
 	unsigned int i;
 
-	if (imu != ctx->dummy_ubuf) {
+	if (imu != &dummy_ubuf) {
 		for (i = 0; i < imu->nr_bvecs; i++)
 			unpin_user_page(imu->bvec[i].bv_page);
 		if (imu->acct_pages)
@@ -459,14 +465,14 @@ static int __io_sqe_buffers_update(struct io_ring_ctx *ctx,
 			break;
 
 		i = array_index_nospec(up->offset + done, ctx->nr_user_bufs);
-		if (ctx->user_bufs[i] != ctx->dummy_ubuf) {
+		if (ctx->user_bufs[i] != &dummy_ubuf) {
 			err = io_queue_rsrc_removal(ctx->buf_data, i,
 						    ctx->user_bufs[i]);
 			if (unlikely(err)) {
 				io_buffer_unmap(ctx, &imu);
 				break;
 			}
-			ctx->user_bufs[i] = ctx->dummy_ubuf;
+			ctx->user_bufs[i] = &dummy_ubuf;
 		}
 
 		ctx->user_bufs[i] = imu;
@@ -1077,7 +1083,7 @@ static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
 	int ret, nr_pages, i;
 	struct folio *folio = NULL;
 
-	*pimu = ctx->dummy_ubuf;
+	*pimu = &dummy_ubuf;
 	if (!iov->iov_base)
 		return 0;
 
-- 
2.41.0